ON TORSIONAL RIGIDITY AND PRINCIPAL FREQUENCIES: AN INVITATION TO THE KOHLER-JOBIN REARRANGEMENT TECHNIQUE LORENZO BRASCO Abstract. We generalize to the p−Laplacian ∆p a spectral inequality proved by M.-T. Kohler-Jobin. As a particular case of such a generalization, we obtain a sharp lower bound on the first Dirichlet eigenvalue of ∆p of a set in terms of its p−torsional rigidity. The result is valid in every space dimension, for every 1 < p < 1 and for every open set with finite measure. Moreover, it holds by replacing the first eigenvalue with more general optimal Poincar´e-Sobolev constants. The method of proof is based on a generalization of the rearrangement technique introduced by Kohler-Jobin. Contents 1. Introduction 2 1.1. Background and motivations 2 1.2. Aim of the paper 3 1.3. Notation 3 1.4. Main result 5 1.5. Plan of the paper 6 2. Preliminaries 6 3. The modified torsional rigidity 10 4. The Kohler-Jobin rearrangement technique 15 5. Proof of Theorem 1.1 19 6. The case of general norms 22 References 27 2010 Mathematics Subject Classification. 35P30, 47A75, 49Q10. Key words and phrases. Torsional rigidity; nonlinear eigenvalue problems; spherical rearrangements. 1 2 BRASCO 1. Introduction N 1.1. Background and motivations. Given an open set Ω ⊂ R with finite measure, we consider the following quantities Z Z 2 jruj2 dx juj dx Ω Ω λ(Ω) = min Z and T (Ω) = max Z : u2W 1;2(Ω)nf0g 2 u2W 1;2(Ω)nf0g 2 0 juj dx 0 jruj dx Ω Ω The first one is called principal frequency of Ω and the second one is its torsional rigidity. Our terminology is a little bit improper, since the usual definition of torsional rigidity differs from our by a multiplicative factor. Since this factor will have no bearing in the whole discussion, we will forget about it. Another frequently used terminology for λ(Ω) is first eigenvalue of the Dirichlet-Laplacian. Indeed λ(Ω) coincides with the smallest real number λ such that the problem −∆u = λ u in Ω; u = 0; on @Ω; has a nontrivial solution1. In [24] P´olya and Szeg}oconjectured that the ball should have the following isoperimetric-type property: (?) among sets with given torsional rigidity, balls minimize the principal frequency: In other words, by taking advantage of the fact that λ(t Ω) = t−2 λ(Ω) and T (t Ω) = tN+2 T (Ω); t > 0; they conjectured the validity of the following scaling invariant inequality 2 2 (1.1) T (Ω) N+2 λ(Ω) ≥ T (B) N+2 λ(B); where B is any ball. We recall that among sets with given volume, balls were already known to minimize λ (the celebrated Faber-Krahn inequality) and maximize T (the so-called Saint- Venant Theorem). This means that the inequality conjectured by P´olya and Szeg}o was not a trivial consequence of existing inequalities. A proof of (1.1) was finally given by Kohler- Jobin in [19, 20], by using a sophisticated new rearrangement technique. The latter is 1;2 indeed a general result which permits, given Ω and a smooth positive function u 2 W0 (Ω), to construct a ball B having smaller torsional rigidity and a radially symmetric decreasing ∗ 1;2 function u 2 W0 (B) such that Z Z Z Z jruj2 dx = jru∗j2 dx and jujq dx ≤ ju∗jq dx; Ω B Ω B for every q > 1. It is clear that once we have this result, the P´olya-Szeg}oconjecture is easily proven. Of course this also shows that (?) is still true if we replace the principal frequency 1 1;2 Here solutions are always understood in the energy sense, i.e. u 2 W0 (Ω) and is a weak (then 1;2 classical if @Ω is smooth enough) solution. It is well-known that by dropping the assumption u 2 W0 (Ω) strange phenomena can be observed, like for example nontrivial harmonic functions being constantly 0 at the boundary @Ω. ON TORSIONAL RIGIDITY AND PRINCIPAL FREQUENCIES 3 λ(Ω) by any other optimal Poincar´e-Sobolev constant. In other words, balls minimize the quantity Z jruj2 dx Ω ∗ 2 N (1.2) min 2 ; where 1 < q < 2 = ; 1;2 Z q N − 2 u2W0 (Ω)nf0g jujq dx Ω among sets with given torsional rigidity (see [18, Theorem 3]). For some related studies on the quantities (1.2), we also mention the recent paper [6]. 1.2. Aim of the paper. Unfortunately, the Kohler-Jobin's rearrangement technique seems not to be well-known, even among specialists. Then the goal of this paper is twofold: first of all, we try to revitalize interest in her methods and results. Secondly, we will extend the Kohler-Jobin inequality to more general \principal frequencies" associated with the nonlinear p−Laplace operator, defined by p−2 ∆pu = div (jruj ru); and to some anisotropic variants of it (Section 6). The main difficulty of this extension lies in the lack of regularity of solutions to equations involving ∆p, indeed in general these are far from being analytic or C1, as required in [18, 19, 20]. We will show that the Kohler-Jobin technique can be extended to functions enjoying a mild regularity property (see Definition 3.1), which is indeed satisfied by solutions to a wide class of quasilinear equations (see Lemma 3.2). Also, we will simplify some arguments used in [18, 19, 20]. For example, in order to compare the Lq norms of the original function and its rearrangement, we will sistematically use Cavalieri's principle, as it is natural. Finally, we will not require smoothness hypotheses on Ω, which is another difference with the work of Kohler-Jobin. 1.3. Notation. In order to clearly explain the contents of this work and the results here contained, we now proceed to introduce some required notation. N 1;p By Ω ⊂ R we still denote an open set with finite measure, while W0 (Ω) stands for the 1 closure of C0 (Ω) with respect to the norm krukLp(Ω). Throughout the whole paper we will always assume that 1 < p < 1. In this work we will consider the “first eigenvalues" Z jrujp dx Ω (1.3) λp;q(Ω) = min p ; u2W 1;p(Ω)nf0g Z q 0 jujq dx Ω where the exponent q is such that 8 N p > 1 < q < ; if 1 < p < N; < N − p (1.4) > : 1 < q < 1; if p ≥ N: 4 BRASCO Then the quantity λp;q(Ω) is always well-defined, thanks to Sobolev embeddings. Some- times we will also refer to λp;q(Ω) as a principal frequency, in analogy with the linear case. Observe that a minimizer of the previous Rayleigh quotient is a nontrivial solution of p−q q−2 (1.5) −∆pu = λ kukLq(Ω) juj u; in Ω u = 0; on @Ω; with λ = λp;q(Ω). The two terms on both sides of (1.5) have the same homogeneity, then if u is solution, so is t u for every t 2 R. Moreover, it is not difficult to see that if for a certain λ there exists a nontrivial solutions of (1.5), then we must have λ ≥ λp;q(Ω). These considerations justify the name “first eigenvalue" for the quantity λp;q(Ω) (see [14] for a comprehensive study of these nonlinear eigenvalue problems). The principal frequency λp;q obeys the following scaling law N−p− p N λp;q(t Ω) = t q λp;q(Ω); then the general form of the previously mentioned Faber-Krahn inequality is p + p −1 p + p −1 (1.6) jBj N q λp;q(B) ≤ jΩj N q λp;q(Ω); with equality if and only if Ω is a ball. In other words, balls are the unique solutions to the problem minfλp;q(Ω) : jΩj ≤ cg: Properly speaking, the name Faber-Krahn inequality is usually associated with the partic- ular case of p = q in (1.6). Since the proof is exactly the same for all range of admissible p and q, this small abuse is somehow justified. The special limit case q = 1 deserves a distinguished notation, namely we will set Z p jvj dx 1 Ω Tp(Ω) = = max Z : λp;1(Ω) v2W 1;p(Ω)nf0g p 0 jrvj dx Ω In analogy with the case p = 2, we will call it the p−torsional rigidity of the set Ω. Of course, inequality (1.6) can now be written as 1− p −p 1− p −p (1.7) jΩj N Tp(Ω) ≤ jBj N Tp(B): For ease of completeness, we mention that inequalities (1.6) and (1.7) have been recently improved in [4, 15], by means of a quantitative stability estimate. Roughly speaking, this not only says that balls are the unique sets for which equality can hold, but also that sets \almost" achieving the equality are \almost" balls. It is useful to recall that the proof of (1.6) and (1.7) is based on the use of the Schwarz 1;p symmetrization. The latter consists in associating to each positive function u 2 W0 (Ω) a # 1;p # # radially symmetric decreasing function u 2 W0 (Ω ), where Ω is the ball centered at the origin such that jΩ#j = jΩj. The function u# is equimeasurable with u, that is jfx : u(x) > tgj = jfx : u#(x) > tgj; for every t ≥ 0; ON TORSIONAL RIGIDITY AND PRINCIPAL FREQUENCIES 5 # so that kukLq = ku kLq for every q ≥ 1.